Known Participant

Question

Grain and Gaussiam Blut 32 bit

Forum|Forum|4 years ago
November 29, 2021
4 replies
2791 views

It's been so long on feature request that it's no longer a feature, by now it's just a crippling bug. Come on, by now half of our projects are in ACES. AE ACES workflow is surprisingly OK, but it does require us to work in 32bit. Still not having an Add Grain and Gaussian blur working in 32bit is just ridiculous.

Seriously, how much time would it take, 1, 2 days for a single developer??

Just as a reminder:

- add grain doesn't work in 32bit at all

- Gaussian Blur has a bug where it generates negatove picel values - whenever an overbright point gets blurred, it gets a full black (actually negative) halo around it.

These have been reported for over 3 years now, why are they not done yet???

Bug

M

Michael Orwell

Participating Frequently

Apparently, this problem has not been solved even in the latest Beta version.
It is outrageous to leave a rudimentary bug that generates negative pixel values in the most used effects, such as Gaussian Blur, lurking for years.

C

chrisbrearley

Inspiring

Same issue for me. Here's Adobe's version of a Gaussian blur in 2024

P

platplat

Participant

Runninginto this issue now too. Have you found a grain solutoin?

Felix Dames

Inspiring

As much as I hate to post this Add Grain workaround for 32bpc projects because Adobe should finally develop Add Grain for 32 bpc... but here is - it might be helpful for other users:

When working in 32 bpc (float) projects for every effect which is not 32bpc color aware you may encapsulate that effect between two HDR Compander Effects to preserve overbrites (color values above 1.0). Be aware you will loose some color precision when compressing and expanding but it's mostly not noticeable. The higher you set the Gain the more you lose color precision. A Gain value of e.g. 4.0 will preserve overbrite to a maximum value of RGB [4.0,4.0,4.0]. Higher overbrites will be clipped to 4.0. So set the Gain value as low as possible while still preserving your maximum overbrite values.

I did that workaround for professional VFX Shots on non graded raw footage ... No color grading artist later in the pipeline ever complaint or even noticed it! I hope that helps 😉

But now back to Adobe's lazyness that in 2022 not every effect is 32 bpc.

Since over 10 years I complain about this bad situation in AE forums, AE Program Feedback and AE User Voice...

I can't understand why Adobe is able to upgrade every effect to work with MFR but didn't took more time and upgraded every effect to 32bpc ??? If you already opened and updated some old source code for every effect why on earth you didn't make them all 32 bpc in one go???

That is a very old user request...

https://adobe-video.uservoice.com/forums/911311-after-effects/suggestions/39198634-add-grain-in-32bit

https://adobe-video.uservoice.com/forums/911311-after-effects/suggestions/37473934-32-bpc-effects

https://adobe-video.uservoice.com/forums/911311-after-effects/suggestions/34599415-32-bpc-fx-all-of-them

https://adobe-video.uservoice.com/forums/911311-after-effects/suggestions/43236384-update-the-add-grain-effect-to-use-gpu-and-32-bit

I beg you Adobe it's time for 32bpc .... no more comments.

Shebbe

Community Expert

It's a lot better to apply grain in log space. While it's true that you can compress and expand values above 1.0 non destructively with HDR Compander the image state in between isn't really useful especially if you have high values. To preserve specular of 16.0 reduces the entire image immensley if the rest of the image sits around 1.0 and below on average. So grain values get applied on incredibily low values but get blown up afterwards by the reverse process making the image look worse and introduce negative values.

Using a Color Profile Converter to go from an arbitrary known linear color space to the same space in log and back is much more robust. For example ARRI LogC/Linear. It doesn't matter what the source image is because we only care about temporarily converting the transfer function from lin to log and back.

That said, it would be nice to have an updated 32bit Add Grain.

In example below I reduce up to 16. The highlight on the big garlic midleft was around 13. HDR Compander produces redicoulus contrasted grain with negative values. Lin to Log behaves as expected.

HDR Compander

Lin to Log

Shebbe

Community Expert

And to be clear it doesn't just concern ACES it's the new Gaus Blur with GPU support that does this in 32b. The legacy gaus blur does work. Would also love to see this fixed! It's one of the most common needed blurs in compositing!

J

jason_bartell

Community Manager

Hi kaczorefx and Shebbe,

I recommend using Fast Box Blur instead of Gaussian Blur. It will not generate negative RGB nor other issues. The fall-off can be as close to a gaussian as you like by increasing the number of iterations.

Cheers,

Jason

Shebbe

Community Expert

Sure thanks, but why a workaround when the actual issue can be fixed. The new Gaussian Blur is tagged as 32bit compliant when it's not.

Sign up

To post, reply, or follow discussions, please sign in with your Adobe ID.

Sign in to Adobe Community

To post, reply, or follow discussions, please sign in with your Adobe ID.

Scanning file for viruses.

This file cannot be downloaded